Pooling

Overview

A primitive to perform max or average pooling. More…

// structs

struct dnnl_pooling_desc_t;
struct dnnl::pooling_backward;
struct dnnl::pooling_forward;

// global functions

dnnl_status_t DNNL_API dnnl_pooling_forward_desc_init(
    dnnl_pooling_desc_t* pool_desc,
    dnnl_prop_kind_t prop_kind,
    dnnl_alg_kind_t alg_kind,
    const dnnl_memory_desc_t* src_desc,
    const dnnl_memory_desc_t* dst_desc,
    const dnnl_dims_t strides,
    const dnnl_dims_t kernel,
    const dnnl_dims_t padding_l,
    const dnnl_dims_t padding_r
    );

dnnl_status_t DNNL_API dnnl_pooling_backward_desc_init(
    dnnl_pooling_desc_t* pool_desc,
    dnnl_alg_kind_t alg_kind,
    const dnnl_memory_desc_t* diff_src_desc,
    const dnnl_memory_desc_t* diff_dst_desc,
    const dnnl_dims_t strides,
    const dnnl_dims_t kernel,
    const dnnl_dims_t padding_l,
    const dnnl_dims_t padding_r
    );

Detailed Documentation

A primitive to perform max or average pooling.

See also:

Pooling in developer guide

Global Functions

dnnl_status_t DNNL_API dnnl_pooling_forward_desc_init(
    dnnl_pooling_desc_t* pool_desc,
    dnnl_prop_kind_t prop_kind,
    dnnl_alg_kind_t alg_kind,
    const dnnl_memory_desc_t* src_desc,
    const dnnl_memory_desc_t* dst_desc,
    const dnnl_dims_t strides,
    const dnnl_dims_t kernel,
    const dnnl_dims_t padding_l,
    const dnnl_dims_t padding_r
    )

Initializes a descriptor for pooling forward propagation primitive.

Arrays strides, kernel, padding_l, and padding_r contain values for spatial dimensions only and hence must have the same number of elements as there are spatial dimensions. The order of values is the same as in the tensor: depth (for 3D tensors), height (for 3D and 2D tensors), and width.

Parameters:

pool_desc

Output descriptor for a pooling primitive.

prop_kind

Propagation kind. Possible values are dnnl_forward_training and dnnl_forward_inference.

alg_kind

Pooling algorithm kind: either dnnl_pooling_max, dnnl_pooling_avg_include_padding, or dnnl_pooling_avg (same as dnnl_pooling_avg_exclude_padding).

src_desc

Source memory descriptor.

dst_desc

Destination memory descriptor.

strides

Array of strides for spatial dimension.

kernel

Array of kernel spatial dimensions.

padding_l

Array of padding values for low indices for each spatial dimension ([[front,] top,] left).

padding_r

Array of padding values for high indices for each spatial dimension ([[back,] bottom,] right). Can be NULL in which case padding is considered to be symmetrical.

Returns:

dnnl_success on success and a status describing the error otherwise.

dnnl_status_t DNNL_API dnnl_pooling_backward_desc_init(
    dnnl_pooling_desc_t* pool_desc,
    dnnl_alg_kind_t alg_kind,
    const dnnl_memory_desc_t* diff_src_desc,
    const dnnl_memory_desc_t* diff_dst_desc,
    const dnnl_dims_t strides,
    const dnnl_dims_t kernel,
    const dnnl_dims_t padding_l,
    const dnnl_dims_t padding_r
    )

Initializes a descriptor for pooling backward propagation primitive.

Arrays strides, kernel, padding_l, and padding_r contain values for spatial dimensions only and hence must have the same number of elements as there are spatial dimensions. The order of values is the same as in the tensor: depth (for 3D tensors), height (for 3D and 2D tensors), and width.

Parameters:

pool_desc

Output descriptor for a pooling primitive.

alg_kind

Pooling algorithm kind: either dnnl_pooling_max, dnnl_pooling_avg_include_padding, or dnnl_pooling_avg (same as dnnl_pooling_avg_exclude_padding).

diff_src_desc

Diff source memory descriptor.

diff_dst_desc

Diff destination memory descriptor.

strides

Array of strides for spatial dimension.

kernel

Array of kernel spatial dimensions.

padding_l

Array of padding values for low indices for each spatial dimension ([[front,] top,] left).

padding_r

Array of padding values for high indices for each spatial dimension ([[back,] bottom,] right). Can be NULL in which case padding is considered to be symmetrical.

Returns:

dnnl_success on success and a status describing the error otherwise.